Deiko Turns Your Cursor and Voice Into a Prompt for Any Coding Agent

The hardest part of working with an AI coding agent is not writing code. It is describing what is on your screen. You hover over a broken dropdown, glance at a console error, and then type a paragraph trying to explain the spatial relationship between a UI element, a behavior, and what you want changed. The agent gets a paraphrase. You get a response that misses the point. The cycle repeats.

Deiko, a macOS app currently on waitlist, tries to collapse that cycle into three motions. Double-tap Right Option to start. Point at what you mean while you speak. Tap Option to stop. A small floating window called a "coin" appears with cropped screenshots of whatever your cursor was hovering over, a text summary of what you said, and a brief that is ready to drop into Claude Code, Cursor, ChatGPT, or Codex.

How the Interaction Actually Works

The core loop is straightforward. You activate Deiko with a double-tap of the Right Option key in any application on any monitor. The app starts following your cursor and listening to your microphone simultaneously. While you speak, you hover over the UI elements you are describing. Deiko crops and labels each screenshot based on where your cursor was at the moment you paused or spoke about it.

You can also hold Option and drag to lasso a specific region of the screen. This is useful when you want to highlight a particular area rather than whatever your cursor happens to be over. Each thing you point at gets its own cropped screenshot with a label.

When you stop talking and tap Option, the floating coin window appears above every screen. It contains the transcription of what you said, a generated summary, and the collection of screenshots. You review the brief, then drag the coin and drop it into whatever agent you are using. The agent starts working immediately from the real UI context, not from a text description of it.

The app supports any language for voice input, and can handle two languages simultaneously. This matters for developers who work in multilingual environments or who think through problems in one language while writing prompts in another.

The Privacy Model

Deiko keeps screenshots on your Mac. They never leave the machine. Audio is deleted as soon as the text brief is generated. After the initial 30 free minutes, transcription runs entirely on-device with no cloud dependency. If you want cloud transcription for better accuracy with mixed languages, the Pro tier lets you bring your own API key from Sarvam or Groq, meaning your voice data goes directly to your chosen provider rather than through Deiko's servers.

No account is required. A license key pasted from your email lives in your keychain. There is no user profile, no cloud sync of briefs, and no telemetry beyond basic analytics on the website.

What the Coin Actually Contains

The coin is not just a screenshot. It is a structured brief that pairs visual context with natural language instruction. When you drop it into Claude Code or Cursor, the agent receives cropped images of the exact UI elements you pointed at, a transcription of what you said about them, and a summary that ties the two together.

The use case the app highlights is debugging a visual bug. You hover over the broken component, glance at the console output, say what should happen instead, and drop the coin into your agent. The agent starts from the real interface, not from a paragraph that tries to describe a layout problem in words. For agents that accept image input, this is a direct signal boost. The model sees what you see.

Pricing and Availability

The app is currently in waitlist mode. When it launches, it will offer 30 free minutes of narration on install, with on-device transcription continuing indefinitely after that. The Pro tier is priced at $29.99 per year, which works out to $2.50 per month. Pro adds cloud transcription for better mixed-language support, the ability to bring your own Sarvam or Groq API key, and session history with replay.

Installation is a single shell command:

curl -fsSL https://deiko.app/install.sh | sh

The app requires macOS 14 or later and runs on both Apple Silicon and Intel Macs.

Where It Fits in the Agent Toolchain

Deiko is not an agent itself. It is an input layer that sits between you and whatever agent you are already using. The list of compatible targets includes Claude Code, Cursor, ChatGPT, Codex, Gemini, and any text field. The coin is a drag-and-drop object that works with all of them.

The closest comparison is a screenshot-to-agent feature in AgentsRoom, which captures a screen region and attaches it to your next agent message. Deiko adds voice on top of that, creating a combined visual-and-verbal brief in a single motion instead of requiring separate screenshot and dictation steps. Spokenly offers system-wide voice dictation for coding agents but does not capture screenshots or pair them with cursor position.

For developers who spend significant time describing UI bugs, visual regressions, or layout problems to AI agents, the combination of cursor-following screenshots with voice input addresses a real friction point. The 30-minute free tier gives enough time to test whether the interaction model works for your workflow before committing to the annual subscription.